Genetic testing of dogs predicts problem behaviors in clinical and nonclinical samples

Abstract

Background
Very little is known about the etiology of personality and psychiatric disorders. Because the core neurobiology of many such traits is evolutionarily conserved, dogs present a powerful model. We previously reported genome scans of breed averages of ten traits related to fear, anxiety, aggression and social behavior in multiple cohorts of pedigree dogs. As a second phase of that discovery, here we tested the ability of markers at 13 of those loci to predict canine behavior in a community sample of 397 pedigree and mixed-breed dogs with individual-level genotype and phenotype data.

Results
We found support for all markers and loci. By including 122 dogs with veterinary behavioral diagnoses in our cohort, we were able to identify eight loci associated with those diagnoses. Logistic regression models showed subsets of those loci could predict behavioral diagnoses. We corroborated our previous findings that small body size is associated with many problem behaviors and large body size is associated with increased trainability. Children in the home were associated with anxiety traits; illness and other animals in the home with coprophagia; working-dog status with increased energy and separation-related problems; and competitive dogs with increased aggression directed at familiar dogs, but reduced fear directed at humans and unfamiliar dogs. Compared to other dogs, Pit Bull-type dogs were not defined by a set of our markers and were not more aggressive; but they were strongly associated with pulling on the leash. Using severity-threshold models, Pit Bull-type dogs showed reduced risk of owner-directed aggression (75th quantile) and increased risk of dog-directed fear (95th quantile).

Conclusions
Our association analysis in a community sample of pedigree and mixed-breed dogs supports the interbreed mapping. The modeling shows some markers are predictive of behavioral diagnoses. Our findings have broad utility, including for clinical and breeding purposes, but we caution that thorough understanding is necessary for their interpretation and use.

Full text

Zapataetal. BMC Genomics (2022) 23:102
https://doi.org/10.1186/s12864-022-08351-9
RESEARCH
Genetic testing ofdogs predicts problem
behaviors inclinical andnonclinical samples
Isain Zapata1,2*, M. Leanne Lilly1, Meghan E. Herron1, James A. Serpell3 and Carlos E. Alvarez1,4,5*
Abstract
Background: Very little is known about the etiology of personality and psychiatric disorders. Because the core
neurobiology of many such traits is evolutionarily conserved, dogs present a powerful model. We previously reported
genome scans of breed averages of ten traits related to fear, anxiety, aggression and social behavior in multiple
cohorts of pedigree dogs. As a second phase of that discovery, here we tested the ability of markers at 13 of those
loci to predict canine behavior in a community sample of 397 pedigree and mixed-breed dogs with individual-level
genotype and phenotype data.
Results: We found support for all markers and loci. By including 122 dogs with veterinary behavioral diagnoses in
our cohort, we were able to identify eight loci associated with those diagnoses. Logistic regression models showed
subsets of those loci could predict behavioral diagnoses. We corroborated our previous findings that small body size
is associated with many problem behaviors and large body size is associated with increased trainability. Children in
the home were associated with anxiety traits; illness and other animals in the home with coprophagia; working-dog
status with increased energy and separation-related problems; and competitive dogs with increased aggression
directed at familiar dogs, but reduced fear directed at humans and unfamiliar dogs. Compared to other dogs, Pit Bull-
type dogs were not defined by a set of our markers and were not more aggressive; but they were strongly associated
with pulling on the leash. Using severity-threshold models, Pit Bull-type dogs showed reduced risk of owner-directed
aggression (75th quantile) and increased risk of dog-directed fear (95th quantile).
Conclusions: Our association analysis in a community sample of pedigree and mixed-breed dogs supports the inter-
breed mapping. The modeling shows some markers are predictive of behavioral diagnoses. Our findings have broad
utility, including for clinical and breeding purposes, but we caution that thorough understanding is necessary for their
interpretation and use.
Keywords: Behavioral genetics, Clinical behavioral diagnoses, Behavior, Anxiety, Fear, Aggression, Social behavior,
Canine behavior, Canine genetics, Canine translational models, Canine, Pedigree dogs, Mixed breed dogs, C-BARQ,
SNP, Genetic testing, Genetic risk
© The Author(s) 2022. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which
permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the
original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or
other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line
to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this
licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/. The Creative Commons Public Domain Dedication waiver (http:// creat iveco
mmons. org/ publi cdoma in/ zero/1. 0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.
Introduction
ere are an estimated 70-90 M pet dogs in 36.5–42%
of US homes [1, 2]. Because dogs suffer from many of
the same conditions as humans and often receive a high
level of health care, they represent an ideal comparative
and translational animal model [3, 4]. Strong positive
selection in dog domestication and breed development
had the by-effect of vastly relaxed negative selection.
As a result, most complex traits studied to date in dogs
Open Access
*Correspondence: superisain@gmail.com; carlos.
alvarez@nationwidechildrens.org
2 Department of Biomedical Sciences, Rocky Vista University College
of Osteopathic Medicine, Parker, CO 80134, USA
5 Department of Pediatrics, The Ohio State University College of Medicine,
Columbus, OH 43210, USA
Full list of author information is available at the end of the article
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Page 2 of 19
Zapataetal. BMC Genomics (2022) 23:102
– including cancer and other diseases, morphology and
behavior – have shown dramatically-reduced polygenic-
ity and moderate-to-large effect sizes of associated vari-
ants [3, 5–13]. In contrast, common human complex
disease variations generally have small effect sizes and are
not directly useful medically or experimentally.
We previously reported genome wide association stud-
ies (GWASs) of breed averages for fear and aggression
behaviors in multiple cohorts with different breed make-
ups [11, 12]. at was based on behavioral data using
the survey instrument C-BARQ, which was previously
shown to have adequate internal reliability and test-retest
and inter-rater reliabilities [14–17]. MacLean etal. per-
formed similar scans of those and other behaviors in one
shared and one different cohort, but correcting for body
mass in the association analysis [13]. In a study part-
nered with the present work, we compared the scans of
diverse behaviors with and without correction for body
mass. e correlation of body size and behavior in dogs
has thus been observed in behavioral [17–20] and genetic
studies [11, 12]. Based on biological relevance, we previ-
ously argued that behavior-body size correlations are due
to pleiotropy [12] (rather than population structure [21]).
We and others have now strengthened that evidence
greatly [11, 22].
In much of the world, dogs exist as pedigree (strati-
fied by hundreds of breeds), mixed-breed and village dog
subpopulations (Suppl. Text). Pit Bulls, which are among
the most popular dog types in the US, are a special and
controversial case (see Suppl. Text) [23]. Pit Bulls and
several other breeds are believed by some to be particu-
larly aggressive and dangerous to humans. As a result,
breed specific legislation has increasingly restricted the
conditions of ownership of those breeds. e ancestral
UK Staffordshire Bull Terrier was once selected for dog
fighting and it is unclear to what degree that has con-
tinued or whether the breed type should be considered
especially dangerous. Pit Bull refers to a group of breeds,
some of which are registered by the American Kennel
Club (AKC) and others by different institutions in the US
[24]. A recent study by Gunter etal., published in 2018,
of two dog shelters in different US states compared visual
and genetic classification of Pit Bull-type dogs [25]. Shel-
ter staff had a 76% correct call rate for 114 dogs that were
genetically greater than 25% American Staffordshire Ter-
rier (AST). eir false positive rate for 270 non-AST dogs
was 1.5%. Of the total 919 dogs from both shelters, 238
had an AST genetic signature (24 and 28%) and the aver-
age AST contents were 39 and 48%. Below 25–38% AST
content, the correct visual calling rate of Pit Bull-type
dogs falls rapidly. A C-BARQ-based behavioral study by
Duffy etal. published in 2008 of ~ 3800 AKC registered
dogs from 32 breeds also included 132 Pit Bull-type
dogs as defined here [17]. Pit Bull-type dogs showed sig-
nificantly decreased aggression to owners, but increased
aggression to dogs (they did not rank highest in any
behavior). Here we mitigated visual calling of Pit Bull-
type dogs by performing principal components analyses
(PCA) with the set of genetic markers genotyped. With
these caveats, this is the first genetic study of Pit Bull
behavior.
As clinical and lay access to genetic testing continues to
accelerate rapidly, it is important to understand its utility.
In order for genetic tests to be clinically actionable, they
have to be useful in the observation, diagnosis or treat-
ment of patients. Knowledge of increased genetic risk can
indicate therapeutic intervention, initiation and inter-
pretation of disease screening, and life planning [26]. In
domesticated animals, the latter includes management of
commercial/production traits, welfare and reproductive
planning. Because complex traits in domesticated species
often have greatly reduced polygenicity and increased
effect sizes of variations compared to humans, the util-
ity of genetic testing in veterinary medicine and animal
sciences is greatly simplified. Our long-term goal is the
development and validation of genetic testing for use by
veterinary behaviorists as well as dog breeders, shelter
administrators and owners.
e objective of this work was to provide further evi-
dence for the previous interbreed findings in a com-
munity sample. Whereas our GWASs were performed
using breed averages of C-BARQ traits and unrelated
genotyped-cohorts with varied breed makeups, here we
used individual-level C-BARQ phenotypes and geno-
types for 20 markers at 13 behavioral GWA loci. Our
400-dog cohort included a typical proportion of pedi-
gree and mixed-breed dogs for the US, and was repre-
sentative of the community and the veterinary behavioral
clinic. Only variations common across breeds could have
been mapped with our approach and such variations are
enriched for admixture [27]. at is consistent with our
present results because correlations between unlinked
markers (associated with population structure) and
between behaviors were distinct. Our findings lend sup-
port for the genome scans and utility of genetic testing,
but, in the Discussion, we advise caution on direct-to-
consumer tests.
Methods
Study design, cohort andgenotyping
Previous GWA discovery was performed using breed
averages of behavior and unrelated genotyped cohorts
of diverse pedigree dogs. In contrast, the present study i)
targeted a subset of those GWA loci; ii) used both pedi-
gree and mixed breed dogs; iii) used dogs with individ-
ual-level phenotypes and genotypes; and iv) included the
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Page 3 of 19
Zapataetal. BMC Genomics (2022) 23:102
original behavioral traits and additional ones. Factor-
ing the complexities of the quantitative and population
genetics, and our power, this work is a second phase of
discovery – with the GWASs being the first.
We designed our study to evaluate the performance
of genetic markers as predictors of canine problematic
behavior in the community. We recruited subjects with-
out breed or geographical restrictions (Suppl. Text). Dog
clinical background and demographic data were provided
by owners in the form of paper questionnaires, while
behavioral information was collected via electronic ques-
tionnaires (C-BARQ). Paper questionnaire and genotype
data were considered predictor variables and C-BARQ
traits were considered response variables.
Our dog cohort included a total of 397 dog subjects.
Descriptive statistics of our sample are provided in
Table1. Our sample had an almost even female to male
ratio (47:52%) and most were neutered (365 vs. 31). All
dogs were considered pets, 16 were classified as working-
purposed and 17 as competition-purposed (2 as both).
45% of our dogs were members of 77 pedigree or designer
breeds (Suppl. TableS1) and 55% were mixed breed. is
is similar to the US proportion of mixed-breed dogs of
51–53% [1, 28]. Owners were asked to describe the
breed make-up of their dog. We evaluated popularities
of breeds in the US and in select US cities, and deter-
mined our cohort to be representative of a typical US
community despite being geographically biased for Ohio
(Suppl. TableS2). Owners most commonly acquired dogs
in our study from shelters, breeders and acquaintances.
Other sources were pet stores and rescue organizations.
Many dogs previously had other owners (e.g., most shel-
ter dogs). Our cohort was intentionally biased to have
increased representation of dogs with a behavioral diag-
nosis: 30% of our sample, of which 21% of those (or 6.5%
of all dogs) were under medication for problem behav-
ior. 30% of our subjects had a non-behavioral medical
condition (allergic, orthopedic, digestive, dermatologic,
ophthalmic,…). Lastly, we noted whether dogs lived with
other dogs, animals or children.
We classified dogs reported to be Pit Bull or Stafford-
shire Terrier/AST as Pit Bull-type dogs, which made up
15% of our cohort. e Principal Components Analyses
(PCA) reported below and further evidence discussed in
the Supplementary Text show patterns that are consist-
ent with known Pit Bull classification rates and percent
breed makeups [25]. Owners provided dog cheek swabs
for DNA isolation. We used custom TaqMan™ quan-
titative Polymerase Chain Reaction (qPCR) assays to
genotype SNPs at 20 markers associated with problem
behaviors in our mapping studies (Methods; Table2 and
Suppl. TableS3). e markers were taken from the SNP
platforms used in the genome scans and are assumed to
be in linkage disequilibrium (LD) with causal variants in
the tagged risk haplotypes. All allele frequencies, but one,
were in Hardy-Weinberg equilibrium. e Chr1A marker
was detected as two states rather than three and was thus
analyzed as binary. No DNA copy number variant has
been described at this locus, but it remains possible the
binary genotype could reflect the presence and absence
of a structural variant.
Subject recruitment andquestionnaires
Dog owners residing anywhere in the US were recruited
to participate through public announcements. One was
targeted to behaviorally diagnosed dog patients at the
Behavioral Clinic in the Veterinary Medical Center at e
Ohio State University (OSU). Due to regulatory restric-
tions, their medical records were not used here. Internal
announcements to general staff and students were made
at Nationwide Children’s Hospital, the Animal Sciences
Department at OSU and the Blue Buffalo Clinical Trial
Office, Veterinary Medical Center at OSU. Participants
were encouraged to invite other dog owners and to sub-
mit samples from multiple dogs in their household. We
excluded from our study dogs younger than 4 months
old or living with the current owner for less than 1
month. Directly-related dogs (siblings, parents) were
excluded unless the owners indicated they had very dif-
ferent behavior profiles (e.g., if one sibling was behavio-
rally diagnosed but another had no problem behaviors).
We excluded dogs suggestive of aggression during cheek
swabbing (which accounted for a total of one excluded
dog). After a prescreening, a kit was mailed to the address
provided by the participant. is kit included a DNA col-
lection kit (see below), a paper questionnaire to be filled
by the owner about their dog, instructions on how to fill
the C-BARQ online questionnaire, a study consent form
to be signed by the owner, and shipping materials and
prepaid envelope for sending the sample to us. Owners
were instructed to complete the C-BARQ online ques-
tionnaire developed and managed at the University of
Pennsylvania by J.A.S. Only dogs recruited for this study
were used from the C-BARQ data. In addition, a paper
questionnaire (available as Suppl. Data1) was included to
capture additional details (e.g., limited household infor-
mation, and behavioral and medical conditions of dogs).
Subjects with missing information were excluded. Com-
plete participation was compensated with a $5 gift card.
DNA isolation andgenotyping
DNA samples were collected using one Performagene
cheek swab (DNAGenotek Inc. Canada). Samples were
incubated for 4–12 h at 50 °C for nuclease deactivation,
stored at room temperature and processed in batches
following the Performagene PG-AC1 protocol. DNA
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Page 4 of 19
Zapataetal. BMC Genomics (2022) 23:102
Table 1 Descriptive statistics of cohort
Variable Frequency Percent
Sex
Female 189 47.61
Male 208 52.39
Neuter
Fixed 365 92.17
Intact 31 7.83
Working
No 381 95.97
Yes 16 4.03
Compete
No 380 95.72
Yes 17 4.28
Purebred
No 221 55.81
Yes 175 44.19
Pitbull
No 338 85.14
Yes 59 14.86
Acquire Place
Breeder 109 27.53
Other 104 26.26
PetStore 13 3.28
Rescue 25 6.31
Shelter 145 36.62
Other House (lived in a different household)
No 176 55
Yes 144 45
Behavioral Diagnosis
No 275 69.27
Yes 122 30.73
Behavioral Medication
No 371 93.45
Yes 26 6.55
Medical (the dog has a diagnosed medical condition)
No 277 69.77
Yes 120 30.23
Dogs (other dogs present in the haousehold)
No 150 37.78
Yes 247 62.22
Animals (other animals excluding dogs present in the household)
No 200 50.38
Yes 197 49.62
Kids (present in the household)
No 313 79.04
Yes 83 20.96
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Page 5 of 19
Zapataetal. BMC Genomics (2022) 23:102
Table 2 Allele frequencies for sample and diagnosis classes
Marker Full sample No behavior diagnosis With a behavior diagnosis
Frequency % Frequency % Frequency %
Chr1A 6.3 B 7.6 B 3.3 B
A 372 93.7 254 92.4 118 96.7
B 25 6.3 21 7.6 4 3.3
Chr1B 52.6 B 53.6 B 50.4 B
AA 121 30.5 82 29.8 39 32.0
AB 134 33.8 91 33.1 43 35.3
BB 142 35.8 102 37.1 40 32.8
Chr5 31.9 B 32.4 B 30.7 B
AA 193 48.6 135 49.1 58 47.5
AB 155 39.0 102 37.1 53 43.4
BB 49 12.3 38 13.8 11 9.0
Chr10A 11.2 B 10.0 B 13.9 B
AA 321 80.9 229 83.3 92 75.4
AB 63 15.9 37 13.5 26 21.3
BB 13 3.3 9 3.3 4 3.3
Chr10B 71.8 B 70.0 B 75.8 B
AA 59 14.9 43 15.6 16 13.1
AB 106 26.7 79 28.7 27 22.1
BB 232 58.4 153 55.6 79 64.8
Chr10C 44.3 B 42.0 B 49.6 B
AA 151 38.0 116 42.2 35 28.7
AB 140 35.3 87 31.6 53 43.4
BB 106 26.7 72 26.2 34 27.9
Chr10D 57.9 B 59.5 B 54.5 B
AA 92 23.2 63 22.9 29 23.8
AB 150 37.8 97 35.3 53 43.4
BB 155 39.0 115 41.8 40 32.8
Chr10E 57.1 B 53.6 B 64.8 B
AA 105 26.5 87 31.6 18 14.8
AB 131 33.0 81 29.5 50 41.0
BB 161 40.6 107 38.9 54 44.3
Chr13 10.2 B 9.6 B 11.5 B
AA 343 86.4 241 87.6 102 83.6
AB 27 6.8 15 5.5 12 9.8
BB 27 6.8 19 6.9 8 6.6
Chr15A 50.9 B 52.2 B 48.0 B
AA 132 33.3 87 31.6 45 36.9
AB 126 31.7 89 32.4 37 30.3
BB 139 35.0 99 36.0 40 32.8
Chr15B 66.6 B 68.0 B 63.5 B
AA 71 17.9 44 16.0 27 22.1
AB 123 31.0 88 32.0 35 28.7
BB 203 51.1 143 52.0 60 49.2
Chr18 15.6 B 16.4 B 13.9 B
AA 302 76.1 205 74.6 97 79.5
AB 66 16.6 50 18.2 16 13.1
BB 29 7.3 20 7.3 9 7.4
Chr20 71.2 B 71.5 B 70.5 B
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Page 6 of 19
Zapataetal. BMC Genomics (2022) 23:102
concentrations were determined using a NanoDrop spec-
trophotometer (ermo Fisher Scientific).
We previously reported canine interbreed behavioral
GWASs [12]. at was achieved using C-BARQ breed
stereotypes of behavior and two genome wide SNP
genotype datasets [7, 10]. ose studies were expanded
to include other C-BARQ traits and to add a third SNP
genotype dataset [8] (manuscript in preparation). For
the present work, we selected 20 of those SNP markers
for follow-up and modeling (Table2; Suppl. Table S3).
e loci were primarily selected for veterinary clinical
relevance and prioritized by GWA detection in multi-
ple cohorts (8 were present in 3 cohorts, 8 in 2 and 4 in
1). e latter four were selected for the biochemical or
biological relevance of candidate genes. Some loci have
single markers and others multiple. Most of the latter are
commonly in LD across breeds, but GWA risk alleles at
the second chr10 locus (B-E) and the X locus can be pre-
sent on the same or different haplotypes depending on
the breed (discussed in[12]; [7, 10]). Because the three
GWA cohorts were not genotyped on the same SNP plat-
form, we selected the present markers from the dataset
with the highest resolution at each locus. ese markers
were genotyped using custom TaqMan™ qPCR genotyp-
ing assays manufactured by Applied Biosystems (ermo
Fisher Scientific). Probes were designed using their pro-
prietary probe design tool using sequences from the
CanFam3.1 UCSC Genome browser and considering any
other adjacent SNPs included at the CanFam3.1 assem-
bly included in the Broad Improved Canine Annotation
Table 2 (continued)
Marker Full sample No behavior diagnosis With a behavior diagnosis
Frequency % Frequency % Frequency %
AA 57 14.4 38 13.8 19 15.6
AB 115 29.0 81 29.5 34 27.9
BB 225 56.7 156 56.7 69 56.6
Chr24A 45.2 B 44.5 B 46.7 B
AA 160 40.6 111 40.7 49 40.5
AB 112 28.4 81 29.7 31 25.6
BB 122 31.0 81 29.7 41 33.9
Chr24B 58.6 B 58.7 B 58.2 B
AA 105 26.5 71 25.8 34 27.9
AB 119 30.0 85 30.9 34 27.9
BB 173 43.6 119 43.3 54 44.3
Chr32 30.0 B 30.4 B 29.1 B
AA 218 54.9 152 55.3 66 54.1
AB 120 30.2 79 28.7 41 33.6
BB 59 14.9 44 16.0 15 12.3
Chr34 18.9 B 19.3 B 18.0 B
AA 271 68.3 187 68.0 84 68.9
AB 102 25.7 70 25.5 32 26.2
BB 24 6.1 18 6.6 6 4.9
ChrXA 54.8 B 58.0 B 47.5 B
AA or A 156 39.3 97 35.3 59 48.4
AB 47 11.8 37 13.5 10 8.2
BB or B 194 48.9 141 51.3 53 43.4
ChrXB 53.3 B 55.6 B 48.0 B
AA or A 165 41.6 108 39.3 57 46.7
AB 41 10.3 28 10.2 13 10.7
BB or B 191 48.1 139 50.6 52 42.6
ChrXC 37.7 B 41.5 B 29.1 B
AA or A 222 55.9 142 51.6 80 65.6
AB 51 12.9 38 13.8 13 10.7
BB or B 124 31.2 95 34.6 29 23.8
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Page 7 of 19
Zapataetal. BMC Genomics (2022) 23:102
v.1 [29]. TaqPath ProAmp Master mix was used. Assay
conditions were optimized and qPCR assays were run on
96 well plates on an Applied Biosystems 7500 Real Time
PCR instrument using the standard protocol. Genotype
data are available as Supplementary Data1.
Statistical analysis
Descriptive statistics forcorrelation studies andPCA analysis
All statistical analyses in this work are reported using
the CanFam3 nomenclature for SNP alleles: Reference
is A and Alternative is B. e analyses were performed
on SAS Enterprise Guide v.7.1 (SAS Institute, Cary, NC)
running base SAS v.9.4 and SAS/STAT v.14.1. Variables
included in this study were of three types: continuous
variables (Suppl. TableS4), binary variables and multi-
level categorical variables. Each of those types of vari-
able has its inherent properties which were evaluated and
analyzed on a case by case basis. No data transformations
were necessary or implemented. Descriptive statistics
were calculated using PROC MEANS for continuous
variables and PROC FREQ for binary and categorical
variables and PROC MIXED for combinations of binary/
categorical and continuous variables. Correlations were
calculated using PROC CORR for continuous variables,
and PROC FREQ for binary and categorical variables.
PCA was performed on the genetic markers by assum-
ing a linear dosage effect of the alternate allele and to
C-BARQ traits by assuming a linear dose response of the
alternate B allele. All PCA were performed using PROC
PRINCOMP. Observations with missing values were
omitted in the PCA (but not in the modeling).
Association andstatistical modeling
Association models for behavior, medication and type
of behavioral diagnosis were performed using PROC
LOGISTIC using a full model which included all genetic
markers entered as categorical variables. Behavior, medi-
cation type and behavioral diagnosis modeling were per-
formed only in the subset of subjects that had a formal
diagnosis and those that were medicated within that
subset.
Association models for C-BARQ traits and all ques-
tionnaire and genetic markers were estimated using
PROC MIXED in two modes using all subjects. One
mode included all predictors as a full model mode
(FMM) and a second mode evaluating each predictor
as an individual model mode (IMM). We estimated the
Least Square Means for the “AcquirePlace” multilevel
categorical variable only when it was detected as signifi-
cant. We used Least Square Mean differences to deter-
mine effect directions. Effect directions were reversed
for the Trainability C-BARQ trait because it is the only
variable that captures a positive trait. To perform the
fixed threshold case/control modeling mode (FTCCM),
we used quantile values estimated by PROC MEANS
for each C-BARQ trait at 50, 75, 90 and 95 percentiles to
define case control status of our cohort (Suppl. Data2).
e closest score value to the quantile value above was
used as a threshold and all observations with a value
equal or above the threshold were designated as cases.
Stepwise forward selection models were built by PROC
LOGISTIC using a 0.1 threshold to determine predictors
entering and staying in the model. Effects were deter-
mined by the direction of the odds ratio estimates taking
the event “No”, “Intact”, “Female” and the genotype “AA”
as the baseline. We considered the study exploratory and
used familywise multiple testing correction [30, 31]. e
models FMM, IMM and FTCCM, which used the same
variables, each had a different null hypothesis and fam-
ily of tests. Multiple testing correction was thus based on
the number of parameters per trait in each model. is
p-value threshold corresponds to correction for 40 tests
per trait or p≤ 0.00125.
Results
Variable association andcorrelation analyses
We created association tables to evaluate the relation-
ships of all predictor variables. ose include paper
questionnaire variables, descriptive C-BARQ variables
and marker genotypes. Since variables were continuous,
binary or multilevel categorical, we estimated correla-
tions using different methods (Suppl. TableS4). Figure1
shows pairwise associations of all predictor variables
(see Suppl. Fig.S1 for numerical values). Many predict-
able correlations are evident. Dogs with medical ailments
tended to be older. Neuter status was correlated with the
dog function and source. e dog source was strongly
associated with other variables such as the age the dog
was acquired, neuter status, whether the dog lived in
another household, and Pit Bull-type status. We expected
these associations due to the nature of different sources.
We observed correlations between markers on differ-
ent chromosomes, such as 10, 18, 24, 32 and X. is is
presumably due to genetic stratification across breeds,
but it should be noted that the present loci are likely
to be enriched for admixture [27] and possibly selec-
tion in early domestication [11, 12] (may not reflect
vertical breed-relatedness). Supporting this, correlated
markers were not correlated with the same traits. One
consistently-detected association was between genetic
markers and small body size (using weight as a proxy).
We previously reported this and the association of
small size and problem behaviors [11, 12], which are
supported by behavioral [17–20] and other genetic evi-
dence [22]. Pedigree breed and Pit Bull-type represent
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Page 8 of 19
Zapataetal. BMC Genomics (2022) 23:102
reduced variation, and, thus predictably, showed asso-
ciation with subsets of markers.
To test for cohort deviations and overrepresenta-
tions of traits, we estimated C-BARQ trait associa-
tions through correlation analysis (Suppl. Fig.S2). e
findings raised no such concerns about the cohort.
e observed relationships across C-BARQ behavio-
ral traits reinforce what has been described in detail
[16, 19, 32]. For instance, this analysis and our genome
scans showed a strong relationship between fear and
aggression [11, 12]. We interpret this to mean aggres-
sion frequently stems from fear [12], a finding consist-
ent with behavioral studies of canine aggression [17].
Principal components analyses
We carried out PCA of genotypes and C-BARQ
response variables (Fig. 2). PCA allows visual repre-
sentation of association or sampling bias: data points
which cluster together are more similar than those fur-
ther apart. Figure2A shows PCA of genetic markers,
for which 32.4% of the variance is explained within the
first two dimensions. Some markers on the same chro-
mosome clustered closely together due to LD, such as
a group on chromosome 10. One of three markers in a
large X chromosome region is slightly separated from
the others, suggesting the existence of two haplotypes
Fig. 1 Pairwise association of questionnaire variables and genetic markers. Significance test is shown above the diagonal line and effect size and
direction below (odds ratio for categorical variables and estimate ratio for continuous variables). SNP alleles are given according to the CanFam3
nomenclature: Reference allele is A and Alternative is B (A/B should be considered arbitrary assignments without regard to population frequencies
or ancestral/derived status). Genetic marker significance test and correlation are for AA vs BB. In the top right of the diagonal, light red denotes
significant association, p ≤ 0.05; and dark red is significant association, p ≤ 0.001 (for actual p-values, see Suppl. Fig. S1). In the bottom left of the
diagonal, red is positive association and blue negative. Values are colored in a gradient from red to blue according to their value. Only values with a
significant association are displayed
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Page 9 of 19
Zapataetal. BMC Genomics (2022) 23:102
(for demonstration at the present X chromosome locus,
see Figs.6 and 8 in ref. [12]).
Figure2B shows PCA of the C-BARQ response vari-
ables, for which 25% of the variance was explained in the
first two dimensions. is plot is consistent with our pre-
viously reported correlations between fear and aggres-
sion traits [11, 12]. We further evaluated the uniformity
observed in our association analysis among the Pit Bull-
type dogs. To accomplish this, we plotted the PCA data
for genotype and C-BARQ scores, but colored according
to Pit Bull status (Fig.2C/D). A cluster of Pit Bull-type
samples in the lower left side of the plot indicated those
dogs are genetically more homogeneous with each other
and different from the other dogs in the cohort. Examina-
tion of those dogs with pedigree dogs suggests the more
homogeneous dogs are purer Pit Bulls. at is, the more
tightly-clustered these dogs are, the more similar they
are to breeds closely related to Pit Bulls (Suppl. Text). We
similarly considered sex, neuter status, pure pedigree,
mixed breed, behaviorally diagnosed, and other non-
behavioral medical ailments (Suppl. Figs.S3 and S4). Sev-
eral of those plots showed deviations from randomness.
For comparison to Pit Bull-type dogs, we generated plots
for common pedigree breeds in our cohort: the com-
bined members of the retriever group, German Shepherd
Dogs and Rottweilers (Suppl. Fig.S5).
Behavioral diagnosis prediction based ongenetic markers
We next tested whether our set of 20 markers could pre-
dict the risk of dogs having a behavioral diagnosis. We
used logistic regression models to evaluate our cohort
of 397 total dogs, 122 of which had a behavioral diag-
nosis (incl. Diagnosed Pit Bull-type, n= 20 out of 122;
and dogs medicated for behavior, n= 26 out of 122). e
models considered all genetic predictors simultaneously,
thus accounting for (but not estimating) their correla-
tions. e results showed a set of five markers on three
chromosomes (10, 13 and X) can predict a behavioral
Fig. 2 Principal components analysis (first two components). A Genetic markers, B C-BARQ behavioral traits, C, D genetic markers and C-BARQ
behavioral traits, respectively, with Pit Bull-type classification
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Page 10 of 19
Zapataetal. BMC Genomics (2022) 23:102
diagnosis (Fig. 3; p-values given in Suppl. Fig. S6). In
most cases, the loci associated with any behavioral diag-
nosis were different from those associated with a specific
diagnosis. is was possible because the tests of any diag-
nosis were done on the full cohort, but the tests for spe-
cific diagnoses were done on only the subset of dogs with
any diagnosis. e top candidate genes at those loci are
MSRB3 and HMGA2, ANGPT1 and IGSF1, respectively
[12]. (e RSPO2 haplotype associated with canine coat
traits is near, but distinct from, the chr13/ANGPT1 risk
haplotype [11]). ere are multiple haplotypes at the sec-
ond chromosome 10 risk locus (chr10B-E markers) and
those are associated with various morphological and
behavioral traits [7, 10, 12]. Subsets of these traits can
be correlated in some breeds, such as small size, floppy
ears and increased fear, anxiety and aggression. e chro-
mosome 13 risk haplotype is associated with multiple
behavioral traits, including increased fear, anxiety and
aggression traits, as well as smaller size. Lastly, X chro-
mosome locus markers near IGSF1 are associated with
fear, anxiety, aggression and body size traits, and markers
near HS6ST2 are associated with sociability [10–12]. Fur-
ther behavioral analyses are necessary, but the first impli-
cation is that fear, anxiety and aggression are the most
important emotional or personality traits associated with
a clinical diagnosis of problem behaviors.
Genetic markers which are strongly correlated tend to
have redundant effects due to collinearity. However, none
of the marker correlations present in the association
tables and PCA plots reached significance. For example,
the marker chrXC was not significantly associated with
chrXA/B although they clustered relatively closely in the
PCA plot (Fig. 2A). No marker was associated with Pit
Bull-type among behaviorally diagnosed dogs, indicating
the breed is not unique or behaviorally-defined by any
single variant in our panel. e allele of IGF1 (chr15)
that explains the most variance in small body size in dogs
[9, 33] predicted dogs currently on medication for their
behavioral diagnosis.
In terms of specific diagnoses made by a clinical behav-
iorist, we detected three significant marker associations
with anxiety disorder: chr1 near ESR1, chr20 near MITF
and chr24 near RALY, EIF2S2 and ASIP. e relevant
mapped traits were fear of unfamiliar dogs for Chr1A,
separation urination and separation anxiety for Chr20,
and nonsocial fear, touch sensitivity and separation anxi-
ety for Chr24A (Suppl. Table S3) [11, 12]. No marker
was associated with a fear diagnosis but a chr10 marker
between MSRB3 and HMGA2 was associated with diag-
nosis of aggression. e relevant trait mapped for Chr10E
was for aggression toward unfamiliar dogs. We did not
detect associations with compulsive behavior and sleep
disorder diagnoses. is was not surprising because
the frequencies of those diagnoses were very low in our
cohort and, therefore, only very large effects could have
been detected. Our genetic testing results are consistent
with reporting by the veterinary behavioral field [34, 35],
and suggest clinical relevance and, presumably, broader
use.
Statistical modeling ofC‑BARQ behavioral traits
Description ofthree models
We used statistical modeling to determine the relevance
and effect direction of each predictor variable – from
the paper questionnaire and genotype markers – for
each C-BARQ variable. We applied three modes. e
full model mode (FMM) included all predictive vari-
ables together. FMM offers risk estimation incorporating
Fig. 3 Diagnostic prediction. Top shows significant marker prediction of a behavioral diagnosis and medication usage (p-values are given in Suppl.
Fig. S6). Bottom shows significant marker prediction of specific behavioral diagnoses
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Page 11 of 19
Zapataetal. BMC Genomics (2022) 23:102
covariation introduced by other variables in the model.
e individual model mode (IMM) included each pre-
dictive variable individually. e IMM does not take
any covariance into account and offers risk estimation
independent of other predictors. e fixed threshold
case-control mode (FTCCM) stratified risk according to
trait severity, but has increased uncertainty. As severity
increases, there are fewer cases and power decreases.
erefore, each mode has its own inferential application
that requires further evaluations of utility. Due to the sta-
tistical power that can be achieved in our sample, it was
not feasible to include individual breeds in our models.
We believe this is unnecessary because our risk alleles
were mapped by interbreed GWA in multiple cohorts
and are thus common across diverse breeds rather than
representing specific breeds or breed groups [27].
Individual andfull model modes
Overall, the IMM detected more significant associations
than the FMM (Figs.4 and 5; p-values given in Suppl.
Figs. S7/S8). e most consistent predictor variables
were: i) having a behavioral diagnosis (16/36 FMM vs.
15/36 IMM), which consistently increases risk of prob-
lematic behavior for several traits; ii) participating in
competitive sports (10/36 FMM vs. 9/36 IMM), which
reduces risk for most problematic traits except for famil-
iar dog aggression; iii) age of acquisition (9/36 FMM
vs. 6/36 IMM); and iv) age at evaluation (9/36 FMM vs.
11/36 IMM). e latter two are associated with both
reduced and increased risks of different traits. Consistent
with previous behavioral studies [12, 19], we found that
larger dogs are considered to be more trainable. Work-
ing dogs had higher risk of separation-related problems,
increased energy and coprophagia (FMM only). e
energy trait, and possibly separation, is consistent with a
previous study of Swedish military working dog (SMWD)
temperament using C-BARQ phenotypes [36]. SMWDs
which passed temperament tests were more hyperactive/
restless – which we showed here to be correlated with
energy (p< 0.001, Fig. S2) – than those who did not, and
were, on average, left home alone more hours per day in
their first year of life. Dogs with non-behavioral medical
ailments had a reduced risk of displaying many prob-
lematic behaviors. ey also showed an increased risk
of aggression directed at familiar dogs (FMM and IMM)
and coprophagia (IMM only).
Having other non-canine animals in the home was
associated with reduced risk of chasing and increased
risk of coprophagia and hyperactivity (FMM and IMM).
Having other dogs vs. other animals in the home only
overlapped for increased hyperactivity (both modes).
Dog body size had two behavioral trait associations in the
FMM and nine in the IMM. is hints that small body
size is a relatively good predictor when more information
is unavailable, consistent with previous reports [17–19].
Pit Bull-type designation was not predictive of aggressive
behavior, but reduced risk of coprophagia and excitability
(FMM), and increased risk of leash pulling (both modes).
Having children in the home increased the risk of snap-
ping at flies and shadow chasing, and reduced the risk
of stranger-directed fear (IMM) and other stereotypic
behavior (“Other Behaviors”; bizarre, strange, or repeti-
tive behavior). e source of acquisition from was predic-
tive of excitability, some types of aggressive behavior, and
trainability (IMM). Dogs obtained from a shelter or from
a breeder tended to have lower risk of problem behaviors
than dogs from pet stores. at is consistent with the fact
that dogs purchased at pet stores have increased risk of
problem behaviors compared to dogs from non-com-
mercial breeders [37]. is could be confounded by small
body size, which is genetically associated with problem
behaviors [11, 12]. e last two decades have experi-
enced a trend of increased popularity of smaller pedigree
breeds [19, 38, 39].
e most consistent genetic effects on the 36 traits
came from two body size loci [9]: the chr15B marker at
the IGF1 locus (6/36 FMM vs. 12/36 IMM) which always
increased risk, and chr34 near IGF2BP2 (6/36 vs. 6/36),
which, in the majority of cases, increased risk of mul-
tiple problem behaviors. e most consistent genetic
effect predicting fear and aggression was chr18 at the
GNAT3/CD36 locus, as previously reported [11, 12]. In
both FMM and IMM, the chr20 marker near MITF (4/36
FMM vs 8/36 IMM) predicted reduced risks of inappro-
priate chewing and chasing. In both modes, chr10E con-
sistently predicted reduced risk of aggression directed at
unfamiliar dogs, but mixed effects for other traits. Some
markers, like chr32 near RASGEF1B, showed an effect
across several traits but only in a single mode. Chr10B,
chr10D, chr13, chr24A and all three chrX loci were only
significant in IMMs. e C-BARQ trait with the most
genetic predictors was aggression directed at unfamiliar
dogs (6/20 FMM vs. 4/20 IMM). Interestingly, owner-
directed aggression had 5/10 genetic predictors in IMM
but none in FMM. Both modes consistently showed
chr10A near LRIG3 predicted aggression directed at
unfamiliar dogs. In summary, genetic testing consistently
predicted multiple C-BARQ traits, including in the areas
of fear, anxiety and aggression.
Fixed threshold case‑control mode
We used the FTCCM to test the robustness of signifi-
cant predictors and determine how association patterns
are affected as behavioral C-BARQ traits are strati-
fied by severity. We set fixed threshold values for each
C-BARQ trait at the 50th, 75th, 90th and 95th quantile
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Page 12 of 19
Zapataetal. BMC Genomics (2022) 23:102
levels and deemed those above the threshold as cases.
We analyzed the data as a binary response variable.
We created models using logistic regression with a
stepwise forward selection. Overall, FTCCM models
showed a similar pattern as the FMM and IMM gen-
eralized linear models (Suppl. Figs. S9, S10, S11 and
S12). As fixed thresholds were raised with concomi-
tant loss of power, we detected some interesting effects.
Weight, age and age at evaluation decreased their rel-
evance as the threshold increased. As was observed in
FMMs and IMMs, having a behavioral diagnosis con-
sistently predicted problematic behavior and dogs with
Fig. 4 Full Model Mode (FMM). Generalized linear model associations of C-BARQ behavioral traits by questionnaire and genetic markers were
evaluated together. Each behavioral trait was modeled but only significant effects are highlighted. SNP alleles are given according to the
CanFam3 nomenclature: Reference allele is A and Alternative is B (A/B should be considered arbitrary assignments without regard to population
frequencies or ancestral/derived status). Green denotes decreased risk and red increased risk of the A vs. the B allele. A darker shade of green or
red denotes significant at a Bonferroni level adjusted by trait. Actual p-values are given in Supplementary Fig. S7. When the effect of place acquired
(AcquirePlace) is significant, the Least Square Mean estimate of each of its levels is shown in the columns to its right; color gradient is arranged from
lowest to largest
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Page 13 of 19
Zapataetal. BMC Genomics (2022) 23:102
non-behavioral diagnoses had lower risk of some prob-
lem behaviors.
As expected, FTCCM models were the least sensitive.
Genetic marker performance for FTCCM models was
not as robust as in FMMs and IMMs. e exception was
chr5 (near SHISA6), which exhibited the least variability
in the PCA (Fig.2A) and was not significant in the FMM
and IMM. Chr5 was mapped for escaping and chas-
ing [11], and here was associated with escaping in the
FTCCM. ChrXB predicted milder cases of urine mark-
ing. Chr10E, chr18, and chrXC were the most relevant
for fear and aggression traits. Chr10E was most relevant
Fig. 5 Individual Model Mode (IMM). Generalized linear model associations of C-BARQ behavioral traits by questionnaire and genetic markers
were evaluated individually. Each behavioral trait was modeled but only significant effects are highlighted. SNP alleles are given according to the
CanFam3 nomenclature: Reference allele is A and Alternative is B (A/B should be considered arbitrary assignments without regard to population
frequencies or ancestral/derived status). Green denotes decreased risk and red increased risk of the A vs. the B allele. Green denotes decreased
risk and red increased risk. A darker shade of green or red denotes significant at a Bonferroni level adjusted by trait. Actual p-values are given in
Supplementary Fig. S8. When the effect of acquired place (AcquirePlace) is significant, the Least Square Mean estimate of each of its levels is shown
in the columns to its right; color gradient is arranged from lowest to largest
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Page 14 of 19
Zapataetal. BMC Genomics (2022) 23:102
when the threshold was lower (50 and 75th quantiles),
suggesting it segregated milder cases of dog directed fear.
Chr18 and chrXC were most relevant for detecting inter-
mediate cases when thresholds were set to 75-90th quan-
tiles (aggression directed at unfamiliar humans and fear
directed at unfamiliar dogs, respectively). ree mark-
ers had increased relevance for detecting problematic
behavior of greater severity, 90-95th thresholds: chr1A
and chr34 for touch sensitivity, and chr20 for separation-
related defecation. Chr32 was associated with increased
trainability (90-95th thresholds). Curiously, the chr32
trainability association was not present in the FMM and
IMM, but reduced fear of stairs was observed in all three
models and at all FTCCM thresholds. Pit Bull-type dogs
were not associated with any fear or aggression trait in
the FMM or IMM models. In the FTCCM, they showed
reduced risk of owner-directed aggression only at the
75th quantile of severity and increased fear of unfamiliar
dogs only at the 95th (discussed in Suppl. Text).
Discussion
One strength of genome scanning of breed averages
is the ability to map alleles that are fixed in individual
breeds [7, 9, 10, 12]. is can complicate interpretation
and validation in those breeds, but that can be addressed
in other breeds and in mixed breed dogs. A drawback
of the approach is that it cannot detect variants that are
rare across breeds [40, 41]. A second strength of inter-
breed mapping is that causal variants can be assumed to
lie within the minimal overlap region across breeds car-
rying the risk haplotype [7, 9, 10, 12]. Because meiotic
recombination events happen independently in each
breed, LD breaks down on both sides of causal variants
and the markers tagging them. As a result, the peak inter-
vals in interbreed GWASs tend to be much smaller than
in single breed GWASs. Additional virtual fine-mapping
is possible by breed-specific phasing of GWA haplotypes
to further refine the minimally overlapping region [11,
12]. Notably, our original GWASs were made possible
by using crowdsourced C-BARQ phenotypes and unre-
lated genotype datasets of only partially-overlapping dog
breeds. Here we tested 20 SNP markers at 13 of those loci
for behavioral associations in a 397-dog cohort designed
to randomly sample the community and behavioral clinic.
In contrast to the GWASs, the present study had individ-
ual-level C-BARQ phenotypes and genotypes. Because
of the high complexity of this work (incl. 17 behavioral
traits and the use of a cohort with half participants being
mixed-breed dogs and the other half being pedigree dogs
representing 77 breeds) and the low power of the cohort,
we consider the GWASs a first phase of discovery and
this study a second phase. Our findings supported behav-
ioral associations for all loci tested, but confirmatory
studies will require a narrower scope or much greater
power.
Control of population structure is critical to genetic
studies of domesticated species [4, 21, 42]. We previously
mitigated the effects of population structure by using
linear mixed models and multiple cohorts with partially
overlapping breed makeup in the discovery GWASs [11,
12]. We also provided evidence for a subset of markers
through predictive modeling in a third group of dogs
with no breeds overlapping the GWASs [12]. Using many
breeds and multiple cohorts with different breed make-
ups reduces the risk of false positives due to population
structure and latent variables such as cryptic related-
ness and batch effects. Here, we observed correlations
between unlinked markers in our cohort. is is consist-
ent with stratification of genetic variation across breeds.
Despite the large number of breeds included and the
high proportion of mixed breed dogs, this is not surpris-
ing. Breed popularity is so unbalanced that the 10 most
popular breeds accounted for 50% of all 2008 AKC reg-
istrations. We cannot rule out the effects of population
structure on our studies or that some behavioral variants
are part of, or inextricable from, population structure [4,
21, 42]. However, population structure is not a critical
problem here because markers that are correlated geneti-
cally are not correlated with the same traits. For instance,
chr18 and chr34 correlated with several markers associ-
ated with having a behavioral diagnosis; however, chr18
and chr34 did not correlate with behavioral diagnoses
in our association, prediction and modeling analyses
(Figs.1, 3, 4 and 5). e same is true for chr32 and fear
of stairs (see below and Suppl. Text). We considered
our results in the context of high-powered clustering of
breeds according to C-BARQ behavior, which found clus-
ters are most strongly associated with body size, followed
by breed relatedness [39]. PCA classification of our data
according to those clusters showed partial segregation of
our genetic markers but not of behavior (Suppl. Text and
Figs.S13/S14). Lastly, strong biological relevance of can-
didate genes further supports our behavioral associations
[11, 12]. For example, the implicated genes at the two
loci most-associated with fear and aggression directed
at unfamiliar humans and dogs – chr18 and chrX here –
are supported by evidence in rodents of related behaviors
and gene expression in the amygdala to hypothalamic-
pituitary-adrenal axis [12].
is study provided further support for our genome
scans of canine behaviors [11, 12] and suggested their
clinical relevance. Ten markers at eight loci were associ-
ated with having a clinical behavioral diagnosis, and a set
of five of those successfully predicted a diagnosis. Among
the broad corroborating evidence of C-BARQ associa-
tions (Suppl. Text), we found evidence that supports all
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Page 15 of 19
Zapataetal. BMC Genomics (2022) 23:102
four of the original loci replicated in a second cohort
for nine fear and aggression traits (chr10, chr15, chr18
and chrX) [12]. e chr18 and chrX associations in this
cohort support our original interpretation that vari-
ants at those loci are associated with fear and aggres-
sion directed at unfamiliar humans and dogs, but not
with owner-directed aggression [12]. Most of the traits
were further supported by the same trait or a related one
[11, 12]. We also provided further evidence for the GWA
findings for chromosome markers 1B, 10A (very distant
from 10B-E), 20, 24 and 34. While our findings lend sup-
port to many of the mapping results, most variations also
had trait associations that differed from the GWASs [11,
12]. is is unsurprising given the differences in design
and power, and the high levels of pleiotropy known for
human brain traits [43, 44] (discussed below). For exam-
ple, chr32 was associated with aggression in the GWASs
and here. However, that chr32 haplotype differed for
several anxiety traits across the GWASs and here. More
studies are necessary to determine if our GWA of breed
averages or the present study with individual-level data
predicted associations more reliably. We expected this
work to yield more accurate data, but it was recently
shown that GWA of dog body size was dramatically more
powerful using breed averages than individual measures
[45].
Our association analysis of traits (Suppl. Table S4)
revealed several unsurprising associations mentioned in
the Results. Others are potentially more interesting. Mir-
roring trends of American Kennel Club breed populari-
ties in the past two decades [19, 38, 39], we observed dogs
acquired from pet stores tended to be smaller in size.
ere was a negative association between having chil-
dren in the home and having a behavioral diagnosis (dis-
cussed in Suppl. Text). Female sex was associated with
any behavioral diagnosis, whereas males had increased
risk for aggression directed at familiar dogs (see mode-
ling results). ere is no consensus on the effects of sex
on canine anxiety/separation traits. Females are known
to have increased risk of developing fear of unfamiliar
humans and dogs [20, 46]. Intact females have increased
fear of dogs compared to intact males, but levels are
increased further – and the sexes are indistinguishable –
when they are neutered [20]. Males are at increased risk
of being more aggressive than females [46–51]. Here,
neutering of both sexes was positively correlated with
behavioral diagnosis, consistent with previous reports
[20, 51]. e modeling analysis has additional detail for
neuter status, but we do not stress this variable because
only a small percent of our cohort was intact.
Trait correlations should be considered carefully as
they could vary across breeds or be due to environmen-
tal effects. A previous study supports the general negative
correlations of trainability with energy and snapping at
flies we observed [15]. However, there is evidence the
trainability-energy relationship is not fixed. For exam-
ple, in comparisons of breed groups, sighthounds rank at
the top for both trainability and energy [46]. Trainability
and energy can also be positively correlated in working
dogs, but interpretation is complicated due to the effects
of selection of dogs for training and of the training itself
[52, 53]. Among the suggestions of gene-environment
interactions, we found strong correlations of behavioral
phenotypes with presence of children in the home. Lastly,
we note that both barking and coprophagia are more
prevalent in domesticated dogs than wolves [54]. But,
whereas barking seems to be an early target of human
selection, the reason for coprophagia is unknown. Canine
coprophagia generally involves non-autologous fresh
stools [55], which we believe is suggestive of microbiome
inoculation. In humans and mice, transplantation of fecal
microbiota has therapeutic effects on anxiety, depression
and inflammation [56, 57]. Here we found coprophagia
was associated with both behavioral and non-behavio-
ral medical diagnoses. Further studies are necessary to
determine if coprophagia is simply correlated with illness
or if it could be an adaptation with therapeutic benefits.
Our prior [12] and present findings in dogs are sugges-
tive of pleiotropy (Suppl. Text). at is also strongly sup-
ported by comparative genetic analyses of dog behavioral
GWASs [11, 22] and consistent with the rapidly growing
evidence of widespread pleiotropy of behavior in humans
[43, 44]. For instance, we showed risks of many dog
problem behaviors are associated with specific genetic
variants known to cause small body size (IGF1, IGF1R,
IGF2BP2 and HMGA2) and that protection against
problem behaviors is conferred by the large-size IGSF1
haplotype [12]. In the present study, we tested all but
the IGFR1 locus and provided further support for these
relationships. Several canine behavioral traits associated
with reduced body size are correlated with each other
[18], and those effects were consistent with our genome
scanning results [12]. Veterinary behaviorists have previ-
ously shown that small dog size is associated with prob-
lem behaviors [18, 20, 39]. A study of German Shepherds
showed that drug-detection training results in immedi-
ately increased levels of circulating IGF1; and this effect
is potentiated in dogs that have undergone 6 months
of training vs. none [58]. While German Shepherds are
fixed for the non-small body size allele, this finding sug-
gests physiological relevance for trainability – which is
one of the traits we showed to be negatively associated
with the small body size allele of IGF1.
In humans, there are many genetic correlations
between height and psychiatric, behavioral and person-
ality traits, including neuroticism [59], risk tolerance
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Page 16 of 19
Zapataetal. BMC Genomics (2022) 23:102
[60] and smoking cessation [61]. ere is also strong
evidence that body size is associated with differences in
brain structure in humans and dogs, and that those have
functional effects (most commonly reported in the area
of cognition) [11, 62]. Both Insulin/IGF signaling and
downstream pathways (here incl. IGF1, IGF2BP2 and
IGSF1, and HMGA2, respectively) have important roles
in brain development [63, 64]. Presumably correlations of
dog body size and behaviors also involve physiology and
psychology [12, 63]. Our interpretation is that the behav-
ioral genetic pathways we mapped are conserved at least
across mammals. However, although body size has been
under selection in both humans and dogs, the biology
and genetic architecture are dramatically different [11,
65, 66]. Dog body size is mostly explained by a handful
of variations of moderate-to-large effect sizes, whereas
humans have countless variations with weak effects. It
seems likely this would be reflected in the pleiotropy of
those variations.
Our findings for Pit Bull-type dogs have three uncer-
tainties (Suppl. Text). First, the designation of Pit
Bull-type dogs is based on visual appearance and the
expectation that mean AST content was ~ 40–50% [25].
If that AST makeup were correct, we believe our study
of these dogs is justified since the correct classification
rate is 76% for dogs as low as 25% AST [25]. Pit Bull-type
dogs have increased genetic diversity because they rep-
resent multiple breeds and because they are commonly
mixed with other breeds. As a result, true Pit Bull effects
are distorted and diluted, and the power to detect them
is reduced. Secondly, our interbreed behavioral GWASs
could not have identified risk variations that are common
in Pit Bull-type dogs but otherwise rare. And thirdly, our
Pit Bull-type sample is probably under-represented for
dogs know to be exceptionally aggressive or to be very
successful in dog fighting (which would be associated
with increased risk of dog attacks and criminal behav-
ior [23]). Both modes of the generalized linear mod-
eling showed only a single trait association: increased
leash pulling. e FTCCM mode detected decreased
owner-directed aggression at the 75th quantile of sever-
ity and increased unfamiliar dog-directed fear only at
the 95th. A previous study of C-BARQ aggression traits
in approximately 3800 dogs included Pit Bull-type dogs
as defined here [17]. It showed they have reduced risk of
owner directed aggression, as we observed, and increased
risk of aggression directed at dogs – but not humans.
It is unknown whether the latter 11.5% of Pit Bull-type
dogs with increased dog-directed aggression also had
increased fear of dogs. If that were the case, it would
explain our observation of extreme dog-directed fear in a
small subset of this breed type. However, our community
sample of Pit Bull-type dogs showed they are not more
aggressive or more likely to have a behavioral diagnosis
than other dogs. is does not support reliance on breed-
specific legislation to reduce dog bites to humans [23]. As
our genetic findings were restricted to known aggression
variations that have large effect sizes across breeds, it is
necessary to identify and understand the effects of rarer
loci that increase risk of dangerous behavior.
Population structure is the most challenging aspect of
genetics in domesticated species. is can be addressed
by the design of future confirmatory studies in dogs.
ose will also make it possible to measure the propor-
tion of the trait variance explained by single and combi-
nations of variations [13]. We successfully applied such
concepts to canine osteosarcoma, including using the
Intersection Union Test to perform a type of meta-anal-
ysis, modeling polygenic risk within and across breeds,
and validating one breed model in a separate sample
[6]. It is currently not feasible to conduct well-powered
mapping in the several hundreds of existing dog breeds.
However, it is possible to study the most popular breeds
(in the US, 10 breeds account for 50% of all AKC regis-
trations). Alternatively, behavioral genome scans could
be performed in phenotyped mixed-breed dogs [67] and
those haplotypes could be characterized in those and
pedigree dogs. Other factors that are difficult to consider
in this work and should be addressed in follow-on stud-
ies are dog sampling bias, and effects of owner person-
ality and socioeconomics. Here we showed our cohort is
representative of the community. However, we assume
that owners of dogs with behavioral diagnoses have a
higher socioeconomic status than average because they
were recruited through an academic veterinary hospital.
Many canine behavioral variants may require environ-
mental stimuli for a behavioral phenotype to manifest.
Owner personality does not necessarily increase the risk
of owner-directed aggression [68], but owner personal-
ity and psychiatric traits are correlated with increased
rates of fear, anxiety, aggression and other traits [69, 70].
Caution must be used in interpreting the association of
small body size with problem behaviors. Small dogs, as
a group, may have different owner and other environ-
mental characteristics compared to larger dogs (e.g.,
physical and social characteristics of home and neighbor-
hood, amount of time spent alone, and levels of physical
and mental exercise). Especially when experienced early
in life, stress is associated with increased risk of mental
health disorders in humans and dogs [71].
Conclusions
is work provides further support for our interbreed
genome scans of dog behaviors, and expands the rel-
evance to mix-breed dogs. In addition to its utility to
address unmet veterinary needs, there is a strong case
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Page 17 of 19
Zapataetal. BMC Genomics (2022) 23:102
for using dog models to understand human psychiatric
disorders [12, 19, 72–75]. As we previously reported
[12], small body size was associated with many prob-
lem behaviors. e results support our previous find-
ings that fear and aggression traits directed at dogs and
unfamiliar humans cluster together and with non-social
fear [12]. We previously noted that owner-directed
aggression lies outside the latter cluster of traits and
here found evidence suggesting it may be more closely
associated with anxiety traits rather than fear. An
important finding was Pit Bull-type dogs in our com-
munity sample, as a group, were not more aggressive or
likely to have a behavioral diagnosis than other dogs.
As the nascent field of canine behavior advances, it will
be important to better account for human influences on
dog behavior. Our results showed genetic screening of
canine behavior is feasible and suggest it may be useful
for owners, breeders, shelters, working dog institutions
and veterinarians. However, we advise caution with
direct-to-consumer tests until there is a better under-
standing of the behavioral risks associated with these
alleles and others which may be common in few breeds,
but undetectable in our interbreed approach.
Supplementary Information
The online version contains supplementary material available at https:// doi.
org/ 10. 1186/ s12864- 022- 08351-9.
Additional le1: Supplementary Data 1. Paper questionnaire, C-BARQ
phenotype and genotype data. Supplementary Data 2. Cumulative
frequencies used to define Fixed Thresholds for FTCCM. Supplemen‑
tary Text. Extended introduction and discussion. TableS1. Cohort
breed frequencies. TableS2. Comparison of cohort breeds to US and
US cities popularities. TableS3. Study markers list with genome scan
traits. TableS4. Descriptive statistics for continuous variables. Figure S1.
Duplicate of Fig. 1, but with numerical values (incl. p-values). Figure S2.
Correlation table of C-BARQ variables. Figure S3. PCA of C-BARQ traits
isolating the following: sex, neuter status, pedigree vs. mixed breed, Pit
Bull-type, behavioral diagnosis and non-behavioral ailments. Figure S4.
PCA of genetic markers isolating the following: sex, neuter status, pedi-
gree vs. mixed breed, Pit Bull-type, behavioral diagnosis and non-behav-
ioral ailments. Figure S5. PCA of genetic markers separately isolating all
retrievers and German Shepherd Dogs. Figure S6. Duplicate of Fig. 3, but
with p-values. Figure S7. Duplicate of Fig. 4, but with p-values. Figure S8.
Duplicate of Fig. 5, but with p-values. FiguresS9–12. Logistic regression
with stepwise selection modeling with cases classified by trait severity at
50th, 75th, 90th and 95th percentile, respectively. FiguresS13/14. PCA of
genetic markers and C-BARQ behavior isolating Wilson et al. 2018 cluster-
ing of breeds by C-BARQ behavior.
Acknowledgements
We are grateful for the kind participation of owners and the contributions
made by their pets. We thank the Blue Buffalo Clinical Trial Office at The
Ohio State College of Veterinary Medicine for their help towards recruiting
behaviorally diagnosed participants. We thank the Department of Animal Sci-
ences at Ohio State for their help with recruiting participants from the general
community. We thank the Stanton Foundation for making this work possible
through their financial support.
Authors’ contributions
I.Z. and C.E.A. designed the studies. I.Z. designed the statistical analyses. I.Z.
conducted the non-clinical recruitment. M.E.H. and M.L.L. made the veterinary
behavioral diagnoses, and, with I.Z., conducted the clinical recruitment. J.A.S.
processed the C-BARQ questionnaire data and guided its interpretation. I.Z.
performed the experimental work. I.Z. and C.E.A. conducted the analyses and
primary interpretations. C.E.A. and I.Z. wrote the manuscript. All authors were
involved in the final interpretation of the results and contributed to writing
the manuscript. The author(s) read and approved the final manuscript.
Funding
I.Z. was supported by a fellowship from the Stanton Foundation (Cambridge,
MA). C.E.A. was supported by grants from the American Kennel Club CHF
(01660) and the Scottish Deerhound Club of America.
Availability of data and materials
All data are included as Supplementary Information in this work.
Declarations
Ethics approval and consent to participate
All dog samples and information were acquired with informed consent
from dog owners, under an approved IACUC protocol from OSU (Protocol
number 2017A00000116). All methods were performed in accordance with
the relevant guidelines and regulations, including ARRIVE guidelines (Animal
Research: Reporting of In Vivo Experiments). Owner questionnaires were
reviewed by the OSU IRB board and declared exempt. All regulatory require-
ments of the study were approved by the BBCTO at The OSU College of Vet-
erinary Medicine. All laboratory work was performed at The Research Institute
at Nationwide Children’s Hospital, which reviewed the proposed study and
determined it to be IACUC and IRB exempt.
Consent for publication
Not applicable.
Competing interests
The authors declare that they have no competing interests.
Author details
1 Department of Veterinary Clinical Sciences, The Ohio State University College
of Veterinary Medicine, Columbus, OH 43210, USA. 2 Department of Biomedi-
cal Sciences, Rocky Vista University College of Osteopathic Medicine, Parker,
CO 80134, USA. 3 Department of Clinical Sciences & Advanced Medicine,
School of Veterinary Medicine, University of Pennsylvania, Philadelphia, PA
19104, USA. 4 Center for Clinical and Translational Research, The Abigail Wexner
Research Institute at Nationwide Children’s Hospital, Columbus, OH 43205,
USA. 5 Department of Pediatrics, The Ohio State University College of Medicine,
Columbus, OH 43210, USA.
Received: 17 February 2021 Accepted: 25 January 2022
References
1. American Pet Products Association. 2017–2018 APPA national pet owners
survey. americanpetproducts. org [Internet]. APPA. 2018.
2. AVMA. US pet ownership and demographics sourcebook. 2012. p. 1–186.
3. Ostrander EA, Wayne RK, Freedman AH, Davis BW. Demographic history,
selection and functional diversity of the canine genome. Nat Rev Genet.
2017;18(12):705–20.
4. Fenger JM, Rowell JL, Zapata I, Kisseberth WC, London CA, Alvarez CE.
Dog models of naturally occurring cancer. In: Martic-Kehl MI, Mannhold
R, Kubinyi H, Folkers G, editors. Animal models for human cancer, vol. 69.
Weinheim: Wiley-VCH Verlag GmbH & Co. KGaA; 2016. p. 153–221.
5. Alvarez CE. Naturally occurring cancers in dogs: insights for translational
genetics and medicine. ILAR J. 2014;55(1):16–45.
6. Zapata I, Moraes LE, Fiala EM, Zaldivar-Lopez S, Couto CG, Rowell
JL, et al. Risk-modeling of dog osteosarcoma genome scans shows
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Page 18 of 19
Zapataetal. BMC Genomics (2022) 23:102
individuals with Mendelian-level polygenic risk are common. BMC
Genomics. 2019;20(1):1–4.
7. Boyko AR, Quignon P, Li L, Schoenebeck JJ, Degenhardt JD, Lohmueller
KE, et al. A simple genetic architecture underlies morphological varia-
tion in dogs. PLoS Biol. 2010;8(8):e1000451.
8. Hayward JJ, Castelhano MG, Oliveira KC, Corey E, Balkman C, Baxter TL,
et al. Complex disease and phenotype mapping in the domestic dog.
Nat Commun. 2016;7:10460.
9. Jones P, Chase K, Martin A, Davern P, Ostrander EA, Lark KG. Single-
nucleotide-polymorphism-based association mapping of dog stereo-
types. Genetics. 2008;179(2):1033–44.
10. Vaysse A, Ratnakumar A, Derrien T, Axelsson E, Rosengren Pielberg G,
Sigurdsson S, et al. Identification of genomic regions associated with
phenotypic variation between dog breeds using selection mapping.
PLoS Genet. 2011;7(10):e1002316.
11. Zapata I, Hecht EE, Serpell JA, Alvarez CE. Genome scans of dog
behavior implicate a gene network underlying psychopathology in
mammals, including humans: bioRxiv; 2020. https:// doi. org/ 10. 1101/
2020. 07. 19. 211078 (non-reviewed preprint).
12. Zapata I, Serpell JA, Alvarez CE. Genetic mapping of canine fear and
aggression. BMC Genomics. 2016;17:572.
13. MacLean EL, Snyder-Mackler N, vonHoldt BM, Serpell JA. Highly herit-
able and functionally relevant breed differences in dog behaviour. Proc
Roy Soc Biol Sci. 1912;2019(286):20190716.
14. Barnard S, Siracusa C, Reisner I, Valsecchi P, Serpell JA. Validity of model
devices used to assess canine temperament in behavioral tests. Appl
Anim Behav Sci. 2012;138(1–2):79–87.
15. Duffy DL, Serpell JA. Predictive validity of a method for evaluating
temperament in young guide and service dogs. Appl Anim Behav Sci.
2012;138(1–2):99–109.
16. Hsu YY, Serpell JA. Development and validation of a questionnaire for
measuring behavior and temperament traits in pet dogs. J Am Vet Med
Assoc. 2003;223(9):1293–+.
17. Duffy DL, Hsu Y, Serpell JA. Breed differences in canine aggression.
Appl Anim Behav Sci. 2008;114(3–4):441–60.
18. McGreevy PD, Georgevsky D, Carrasco J, Valenzuela M, Duffy DL, Serpell
JA. Dog behavior co-varies with height, bodyweight and skull shape.
PLoS One. 2013;8(12):e80529.
19. Serpell J, Duffy D. Dog breeds and their behavior. In: Horowitz A, editor.
Domestic dog cognition and behavior. Berlin: Springer Berlin Heidel-
berg; 2014. p. 31–57.
20. Puurunen J, Hakanen E, Salonen MK, Mikkola S, Sulkama S, Araujo C,
et al. Inadequate socialisation, inactivity, and urban living environ-
ment are associated with social fearfulness in pet dogs. Sci Rep.
2020;10(1):3527.
21. Friedrich J, Talenti A, Arvelius P, Strandberg E, Haskell MJ, Wiener P.
Unravelling selection signatures in a single dog breed suggests recent
selection for morphological and behavioural traits. Genet Genomics
Next. 2020;1(1):e10024.
22. Gnanadesikan GE, Hare B, Snyder-Mackler N, Call J, Kaminski J,
Miklosi A, et al. Breed differences in dog cognition associated with
brain-expressed genes and neurological functions. Integr Comp Biol.
2020;60(4):976–90.
23. Cornwell JM, Alvarez CE, Gordon JC. Dog bites: the causes and litigation.
In: McLay R, Anderson R, editors. Engineering standards for forensic
application. 1st ed: Academic; 2018. p. 558–74.
24. Dickey B. Pit bull: the battle over an American icon. New York: Vintage
Books, A Division of Penguin Random House LLC; 2017.
25. Gunter LM, Barber RT, Wynne CDL. A canine identity crisis: genetic breed
heritage testing of shelter dogs. PLoS One. 2018;13(8):e0202633.
26. Torkamani A, Wineinger NE, Topol EJ. The personal and clinical utility of
polygenic risk scores. Nat Rev Genet. 2018;19(9):581–90.
27. Parker HG, Dreger DL, Rimbault M, Davis BW, Mullen AB, Carpintero-
Ramirez G, et al. Genomic analyses reveal the influence of geographic
origin, migration, and hybridization on modern dog breed development.
Cell Rep. 2017;19(4):697–708.
28. (AHS) AHS: U.S. pet (dog and cat) population fact sheet. 2012.
29. Hoeppner MP, Lundquist A, Pirun M, Meadows JRS, Zamani N, Johnson
J, et al. An improved canine genome and a comprehensive catalogue of
coding genes and non-coding transcripts. PLoS One. 2014;9(3):e91172.
30. Matsunaga M. Familywise error in multiple comparisons: disentangling a
knot through a critique of O’Keefe’s arguments against alpha adjustment.
Commun Methods Meas. 2007;1(4):243–65.
31. Rubin M. Do p values lose their meaning in exploratory analyses? It
depends how you define the familywise error rate. Rev Gen Psychol.
2017;21(3):269–75.
32. Duffy DL, Kruger KA, Serpell JA. Evaluation of a behavioral assessment
tool for dogs relinquished to shelters. Prev Vet Med. 2014;117(3–4):601–9.
33. Rimbault M, Beale HC, Schoenebeck JJ, Hoopes BC, Allen JJ, Kilroy-Glynn
P, et al. Derived variants at six genes explain nearly half of size reduction
in dog breeds. Genome Res. 2013;23(12):1985–95.
34. Bamberger M, Houpt KA. Signalment factors, comorbidity, and trends
in behavior diagnoses in dogs: 1,644 cases (1991-2001). J Am Vet Med
Assoc. 2006;229(10):1591–601.
35. Salonen M, Sulkama S, Mikkola S, Puurunen J, Hakanen E, Tiira K, et al.
Prevalence, comorbidity, and breed differences in canine anxiety in
13,700 Finnish pet dogs. Sci Rep. 2020;10(1):2962.
36. Foyer P, Bjällerhag N, Wilsson E, Jensen P. Behaviour and experiences of
dogs during the first year of life predict the outcome in a later tempera-
ment test. Appl Anim Behav Sci. 2014;155:93–100.
37. McMillan FD, Serpell JA, Duffy DL, Masaoud E, Dohoo IR. Differences in
behavioral characteristics between dogs obtained as puppies from pet
stores and those obtained from noncommercial breeders. J Am Vet Med
Assoc. 2013;242(10):1359–63.
38. Bump P. America’s favorite dog breeds for the past two decades, ranked.
Washington (D.C.): The Washington Post; 2015.
39. Wilson B, Serpell J, Herzog H, McGreevy P. Prevailing clusters of canine
behavioural traits in historical US demand for dog breeds (1926(−)2005).
Animals. 2018;8(11):197.
40. Ilska J, Haskell MJ, Blott SC, Sanchez-Molano E, Polgar Z, Lofgren SE,
et al. Genetic characterization of dog personality traits. Genetics.
2017;206(2):1101–11.
41. Persson ME, Wright D, Roth LS, Batakis P, Jensen P. Genomic regions asso-
ciated with interspecies communication in dogs contain genes related to
human social disorders. Sci Rep. 2016;6:33439.
42. Turcsan B, Tatrai K, Petro E, Topal J, Balogh L, Egyed B, et al. Comparison
of behavior and genetic structure in populations of family and kenneled
beagles. Front Vet Sci. 2020;7:183.
43. Watanabe K, Stringer S, Frei O, Umicevic Mirkov M, de Leeuw C, Polder-
man TJC, et al. A global overview of pleiotropy and genetic architecture
in complex traits. Nat Genet. 2019;51(9):1339–48.
44. Smoller JW, Andreassen OA, Edenberg HJ, Faraone SV, Glatt SJ, Kendler KS.
Psychiatric genetics and the structure of psychopathology. Mol Psychia-
try. 2019;24(3):409–20.
45. Hayward JJ, White ME, Boyle M, Shannon LM, Casal ML, Castelhano
MG, et al. Imputation of canine genotype array data using 365 whole-
genome sequences improves power of genome-wide association stud-
ies. PLoS Genet. 2019;15(9):e1008003.
46. Tonoike A, Nagasawa M, Mogi K, Serpell JA, Ohtsuki H, Kikusui T. Com-
parison of owner-reported behavioral characteristics among genetically
clustered breeds of dog (Canis familiaris). Sci Rep. 2015;5:17710.
47. Borchelt PL. Aggressive behavior of dogs kept as companion animals:
classification and influence of sex, reproductive status and breed. Appl
Anim Ethol. 1983;10(1):45–61.
48. Guy NC, Luescher UA, Dohoo SE, Spangler E, Miller JB, Dohoo IR, et al.
Demographic and aggressive characteristics of dogs in a general veteri-
nary caseload. Appl Anim Behav Sci. 2001;74(1):15–28.
49. Hsu Y, Sun L. Factors associated with aggressive responses in pet dogs.
Appl Anim Behav Sci. 2010;123(3):108–23.
50. Flint HE, Coe JB, Serpell JA, Pearl DL, Niel L. Risk factors associated with
stranger-directed aggression in domestic dogs. Appl Anim Behav Sci.
2017;197:45–54.
51. Farhoody P, Mallawaarachchi I, Tarwater PM, Serpell JA, Duffy DL, Zink C.
Aggression toward familiar people, strangers, and conspecifics in gona-
dectomized and intact dogs. Front Vet Sci. 2018;5:18.
52. Early JB, Arnott EA, Mascord LJ, van Rooy D, McGreevy PD, Wade CM.
Work-type influences perceived livestock herding success in Australian
working kelpies. Canine Genet Epidemiol. 2018;5:5.
53. Hare E, Kelsey KM, Serpell JA, Otto CM. Behavior differences between
search-and-rescue and pet dogs. Front Vet Sci. 2018;5:118.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Page 19 of 19
Zapataetal. BMC Genomics (2022) 23:102
•
fast, convenient online submission
•
thorough peer review by experienced researchers in your field
•
rapid publication on acceptance
•
support for research data, including large and complex data types
•
gold Open Access which fosters wider collaboration and increased citations
maximum visibility for your research: over 100M website views per year
•
At BMC, research is always in progress.
Learn more biomedcentral.com/submissions
Ready to submit your research
Ready to submit your research
? Choose BMC and benefit from:
? Choose BMC and benefit from:
54. Miklósi Á. Dog behaviour, evolution, and cognition. Oxford: Oxford Uni-
versity Press; 2014.
55. Hart BL, Hart LA, Thigpen AP, Tran A, Bain MJ. The paradox of canine
conspecific coprophagy. Vet Med Sci. 2018;4(2):106–14.
56. Soto M, Herzog C, Pacheco JA, Fujisaka S, Bullock K, Clish CB, et al. Gut
microbiota modulate neurobehavior through changes in brain insulin
sensitivity and metabolism. Mol Psychiatry. 2018;23(12):2287–301.
57. Sylvia KE, Demas GE. A gut feeling: microbiome-brain-immune interac-
tions modulate social and affective behaviors. Horm Behav. 2018;99:41–9.
58. Guelfi G, Casano AB, Menchetti L, Bellicci M, Suvieri C, Moscati L, et al. A
cross-talk between blood-cell neuroplasticity-related genes and environ-
mental enrichment in working dogs. Sci Rep. 2019;9(1):6910.
59. Nagel M, Jansen PR, Stringer S, Watanabe K, de Leeuw CA, Bryois J, et al.
Meta-analysis of genome-wide association studies for neuroticism in
449,484 individuals identifies novel genetic loci and pathways. Nat Genet.
2018;50(7):920–7.
60. Karlsson Linnér R, Biroli P, Kong E, Meddens SFW, Wedow R, Fontana
MA, et al. Genome-wide association analyses of risk tolerance and risky
behaviors in over 1 million individuals identify hundreds of loci and
shared genetic influences. Nat Genet. 2019;51(2):245–57.
61. Liu M, Jiang Y, Wedow R, Li Y, Brazel DM, Chen F, et al. Association studies
of up to 1.2 million individuals yield new insights into the genetic etiol-
ogy of tobacco and alcohol use. Nat Genet. 2019;51(2):237–44.
62. Vuoksimaa E, Panizzon MS, Franz CE, Fennema-Notestine C, Hagler DJ Jr,
Lyons MJ, et al. Brain structure mediates the association between height
and cognitive ability. Brain Struct Funct. 2018;223(7):3487–94.
63. Fernandez AM, Torres-Aleman I. The many faces of insulin-like peptide
signalling in the brain. Nat Rev Neurosci. 2012;13(4):225–39.
64. Mir S, Cai W, Carlson SW, Saatman KE, Andres DA. IGF-1 mediated neuro-
genesis involves a novel RIT1/Akt/Sox2 Cascade. Sci Rep. 2017;7(1):3283.
65. Joshi PK, Esko T, Mattsson H, Eklund N, Gandin I, Nutile T, et al. Directional
dominance on stature and cognition in diverse human populations.
Nature. 2015;523(7561):459–62.
66. von Hinke Kessler Scholder S, Davey Smith G, Lawlor DA, Propper C,
Windmeijer F. Child height, health and human capital: evidence using
genetic markers. Eur Econ Rev. 2013;57:1–22.
67. Holmes B. Not just a hound dog. New Sci. 2017;233(3117):38–41.
68. Dodman NH, Donnelly R, Shuster L, Mertens P, Rand W, Miczek K. Use of
fluoxetine to treat dominance aggression in dogs. J Am Vet Med Assoc.
1996;209(9):1585–7.
69. Dodman NH, Brown DC, Serpell JA. Associations between owner per-
sonality and psychological status and the prevalence of canine behavior
problems. PLoS One. 2018;13(2):e0192846.
70. Herron ME, Shofer FS, Reisner IR. Survey of the use and outcome of
confrontational and non-confrontational training methods in client-
owned dogs showing undesired behaviors. Appl Anim Behav Sci.
2009;117(1–2):47–54.
71. Dietz L, Arnold A-MK, Goerlich-Jansson VC, Vinke CM. The importance of
early life experiences for the development of behavioural disorders in
domestic dogs. Behaviour. 2018;155(2–3):83–114.
72. Overall KL. Natural animal models of human psychiatric conditions:
assessment of mechanism and validity. Prog Neuro-Psychopharmacol
Biol Psychiatry. 2000;24(5):727–76.
73. Dodman NH, Karlsson EK, Moon-Fanelli A, Galdzicka M, Perloski M, Shus-
ter L, et al. A canine chromosome 7 locus confers compulsive disorder
susceptibility. Mol Psychiatry. 2010;15(1):8–10.
74. Tang R, Noh H, Wang D, Sigurdsson S, Swofford R, Perloski M, et al. Can-
didate genes and functional noncoding variants identified in a canine
model of obsessive-compulsive disorder. Genome Biol. 2014;15(3):R25.
75. Krakauer JW, Ghazanfar AA, Gomez-Marin A, MacIver MA, Poeppel D.
Neuroscience needs behavior: correcting a reductionist bias. Neuron.
2017;93(3):480–90.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in pub-
lished maps and institutional affiliations.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

1.
2.
3.
4.
5.
6.
Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers and authorised users (“Users”), for small-
scale personal, non-commercial use provided that all copyright, trade and service marks and other proprietary notices are maintained. By
accessing, sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of use (“Terms”). For these
purposes, Springer Nature considers academic use (by researchers and students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and conditions, a relevant site licence or a personal
subscription. These Terms will prevail over any conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription
(to the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of the Creative Commons license used will
apply.
We collect and use personal data to provide access to the Springer Nature journal content. We may also use these personal data internally within
ResearchGate and Springer Nature and as agreed share it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not
otherwise disclose your personal data outside the ResearchGate or the Springer Nature group of companies unless we have your permission as
detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial use, it is important to note that Users may
not:
use such content for the purpose of providing other users with access on a regular or large scale basis or as a means to circumvent access
control;
use such content where to do so would be considered a criminal or statutory offence in any jurisdiction, or gives rise to civil liability, or is
otherwise unlawful;
falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association unless explicitly agreed to by Springer Nature in
writing;
use bots or other automated methods to access the content or redirect messages
override any security feature or exclusionary protocol; or
share the content in order to create substitute for Springer Nature products or services or a systematic database of Springer Nature journal
content.
In line with the restriction against commercial use, Springer Nature does not permit the creation of a product or service that creates revenue,
royalties, rent or income from our content or its inclusion as part of a paid for service or for other commercial gain. Springer Nature journal
content cannot be used for inter-library loans and librarians may not upload Springer Nature journal content on a large scale into their, or any
other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not obligated to publish any information or
content on this website and may remove it or features or functionality at our sole discretion, at any time with or without notice. Springer Nature
may revoke this licence to you at any time and remove access to any copies of the Springer Nature journal content which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or guarantees to Users, either express or implied
with respect to the Springer nature journal content and all parties disclaim and waive any implied warranties or warranties imposed by law,
including merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published by Springer Nature that may be licensed
from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a regular basis or in any other manner not
expressly permitted by these Terms, please contact Springer Nature at
onlineservice@springernature.com